• Oct. 27, 2019, 4:10 a.m.

    Of almost all the concepts Isaac mentions I found the Self-replicating Machine episode the most tantalising. Can we actually make one?

    And it got me thinking, how do you create Mega-Structures, space colonies, mass-produced mirrors or spacescrapers? You need to create a machine to build it for you, but not just any machine: a machine that makes the machine to make it for you.

    Or a machine that makes the machine that then makes another in order to make a machine to make it for you. (You get the picture)

    So this thread I thought we could jot down ideas on the challenges involved, and any solutions you can think of. For instance I would imagine the following problems, just starting with the basics using today's technology:
    - how do you make a machine that makes another one? A 3D printer is the only solution I can think of at the moment. It is conceivable to print one using another.
    - however complexity in current 3D printers is rudimentary. They use single pre-made materials and are assembled by hand. We need one that can be printed simply.
    - and printing electronics is really hard, so your machine cannot 'think' unless it's possible. In order to print electronics you need materials that need mining and refinement processes that are hard to create and difficult to assemble. So perhaps you have a 'brain' like a central computer, that is manufactured using current-day methods, that 'transmits' instructions simply to 3D printed modules. These modules don't need to think then, they just move and perform tasks.
    - they need power. However this could be centralised as well similar to electronics.
    - The modules can assemble more modules using an easily workable material - what material I am not sure of, but it needs to be commonly found, easily configurable using only basic gathering and simple bonding techniques. Any material needed can be gathered by gathering modules, heated by heating modules, crushed by crushing modules, all of which must be able to made by the original material.

    Which then presents the first question and wall to the concept - what material can be used for the above purpose? Does anyone have any ideas? Also feel free to post other challenges/solutions you could think of too.

  • Oct. 27, 2019, 5:54 a.m.

    Nature has shown us a way. A bacteria is a living machine that divides. Design a machine that grows to a certain point and then splits into two.

  • Oct. 29, 2019, 11:31 a.m.

    The issue of materials
    It might be a good idea to have some sort of hierarchy based on the autonomy of our "replicators". Indeed the "what material can be used for the above purposes" makes it sound like we are limited by a rule saying we can only use the materials we have directly on site. It is obviously necessary at some point but definitely not for the grand majority of projects. Remember the early days of the channel when IA gave the example of a way for us to be using all that surplus energy from the growing Dyson swarm. Arguing that we can use it to run industrial sized particle accelerator (e.g. LBNL’s Bevalac) to transmute hydrogen stolen from solar wind or the "surface" of the Sun into the elements we want. And as a bonus we could harness a little bit of the energy produced (up until lead) to lessen the overall power requirement.
    Once we have that working it would be inefficient to limit ourselves to only use the elements accessible on site.
    Simplifying the work of our machines
    The requirements would be to create solar panels (collecting a small portion of the energy) or simple mirrors to use the full extent of solar thermal energy (e.g. Heliostat 1 2 3 4). The energy can be delivered "ready to use" (see power satellite episode from SFIA) or "raw" where the heliostat would be on the construction site, or even just using that raw power to melt/weld/cook stuff. Of course, the two options are not mutually exclusive: a diffraction grating can be used to precisely separate the wavelengths that we can use to run different kind of highly efficient solar cells (each of them having a narrow range of wavelength at high efficiency is interesting) from the rest of it that we could use to simply heat stuff (solar furnace) or even separate the infrared from the few ultraviolet and x-ray and use the latter to catalyze some useful chemical reaction (or even help printing your circuits :)
    Doing that first part would be extremely difficult since it is the first thing we construct from only moon "dust" (or mercury's most likely), but once those material and energy issues are out of the way we can become insanely efficient, creating "replicator" that only replicate and build more stuff without the need to worry about how to obtain some weird elements by mining and refining for hours.
    Merging different options to enhance our efficiency
    I personally think that the solution lies in copying what we already do today: we ultra-specialize. As Migala pointed out with the bio-mecha parallel, having a robot that does absolutely everything by itself in a specific region of space is akin to how human tribes used to work, or how a bacteria reproduce and does stuff... extremely inefficient when compared to our world right now where soon only one country, in one obscure city will be making the absolute best thing in the world and ship it everywhere all around the planet, and that for basically every single product. (only high tech ones obviously not for everyday food)
    I am not entirely convinced the machine-replicator thing is necessarily the most efficient way of doing things, something halfway in between that and what we already do today could be enough. And also we should be getting there progressively instead of trying to directly create a machine replicator thingy e.g. building an autonomous factory of car/bicycle then when feel confident enough to build a Mega-factory that creates other car factories. Then a Mega-factory that creates mining factories which would collect and send the materials to the car factory, relieving us of the burden of supervising mining operation etc.
    I could totally see bio-engineered microorganisms that would lay "eggs"/baby cells to give birth to an enormous amount of specialized cells that simply eat the food that is in front of them and replicate once then dies leaving its exoskeleton/super-hard body behind. Its poopy "waste" could be made out of a copper oxide that another bacteria would then eat and use high energy radiation to process and leave a trail of copper alloy conducting some current (or any other process that would allow us to have a resilient structures that still have some sort of circulatory system (in the broad sense) to power some specific part or help repair should damage occur)
    To guide them we could use a laser to heat up a surface so that bacterias or something would follow the heat and get clustered near it then die. Leaving a trail of "food" which could have very complex patterns (see the circuits link from before). Yes this is dark but so is our "cherished" Mother Nature: thing1 exists then gets eaten alive by thing2 which itself gets gruesomely murdered and eaten by thing3 which then gets cancer etc. ad infinitum. wonderful world right? This trail would then be followed by our specialized hard-shell cells and its numerous successors.
    To some extent, this really is already happening with our own species, as our own teeth (the hard part that is super useful) is actually made my ultra specialized cells that build up your teeth and then all die en masse, leaving us humans with a very efficient tool for crushing/cutting food without having to keep on sustaining the cells that created them.
    Now "nature" is extremely limited, moving from small chemical potential to another, ever so slightly higher (or lower) at each step. It does the job eventually but takes too much time. What i meant by bio-engineering is something that would be billions of years away from bacterias of today. We need those to be able to use powerful energy sources, jump high energy barrier without being too unstable, and allow them to use compounds/alloys/eutectic mixes that nature itself would never create yet would be far superior and useful for efficient, precise and robust constructions.
    I really think that 3D printing and bio-engineering are not mutually exclusive pathways to get to our K2 civilization achievement thingy. However even if all this can most certainly alleviate some problems with heat dissipation, it doesn't eradicate the problem IA talked about in the 3D printing episode about speed/precision/heat dissipation issues. There will still be trade-offs, what matter is to make sure that what we "lose" is something that we don't care about or that is taken care of by another process.
    Albeit just using nature as is, would already be a good thing compared to our pathetically unsustainable agriculture practices of today e.g. regardless of ethics issues, i personally wouldn't mind eating a "steak" made out of protein-making-algae is it means a 1000 other people can also eat as much as they need.
    I'll edit and add a link the corresponding SFIA episodes I mentioned soon. Nice idea for a long term thread to see all related issues summarized to a single place. Thanks Hub.

  • Oct. 29, 2019, 5:51 p.m.

    Glad you started this! I love self-replicating machine concepts too.

    For materials, right now graphene looks the most promising. Depending alone on how you form it, graphene is a:

    Rigid mechanical structure
    Flexible mechanical structure
    Electrical conductor
    Light emitter
    Solar panel
    Transparent panel
    Opaque panel

    In short, if we ever master graphene production we can build just about every part of a 3D printer, which will then print - more graphene parts. And being mostly one element means a minimalistic extraction pipeline.

    There are also silicon and boron alternatives to graphene, in case carbon is in short supply.

    More later I hope...

  • June 6, 2020, 6:17 p.m.

    There's a basic paradox to self-making machines where a device has to be stronger than itself to make itself. There is a workaround, but it has to be respected in any design.

    In a plastic 3D printer for example, you can never make a nozzle that can print another nozzle. If the 'child' nozzle were made of plastic, then it would melt when you squeezed hot plastic through it. If you tried to print the new nozzle from melted steel, then the 'parent' nozzle would have to have a higher melting temperature than steel, so it could never print another one good enough to do the same job.

    It's not just about extrusion. Can a milling machine make a complete copy of itself? No, because you could never carve a cutting bit unless you had another cutting bit of much stronger material.

    In the 1970s NASA figured out a way around this, which was to use two different materials which could form each other, under different conditions. They proposed using metal and plaster molds. A metal mold can form a shape from liquid plaster, and a plaster mold can form a shape from molten metal.

    This might have been the best we could do in the 1970s but I think we can do a lot better now. Plaster needs a lot of water, and disintegrates with very few reuses.

    I think in the long run we'll be better off with 1) an electronic device forming an electric or magnetic field, and 2) the field levitating a flow of ionized material to print electronics, then goto 1) repeat.

    Kind of a Ying and Yang of self creation. I can't create me, but I can create you, and you can create me. Biological life works this way too; DNA prints RNA, which prints enzymes, which in turn copy the DNA, rinse repeat.

    Anyway in my Google quest to see if anyone is on to this, I did turn up an apparently abandoned amateur project called MetallicaRap. While it's not surprising that such a complicated idea would be abandoned, it is surprising that the author did so much legwork on the project before doing so.

    If I were a space-crazy billionaire like Musk or Bezos, I would throw this project a few million dollars (pocket lint to them) and a team of experts to get it off the ground. He/she who owns self-rep will become King of Space.

  • June 7, 2020, 3:32 a.m.

    Yeah this is precisely what I was thinking. Back to the inability for printing heads to mechanically deposit material quickly, I can only think of a device like a 'sewing machine' - where you have a stream of ions (the thread), interacting with a beam (the needle) in a highly coordinated fashion at extreme speeds. You can then make very complicated structures without the limitations of the macroscopic scale.

    Thanks for the link - MetallicaRap is precisely what is needed. I had a chance to meet Elon Musk before but didn't get to (for a different project), but there may be an opportunity soon so will definitely mention it if it happens! :)

    So now all we need is MetallicaRap to 'print itself' including all the materials needed, perhaps with legs and the ability to move around. As Yvonm correctly mentioned though we could have specialised ones that process materials to be fed into a MetallicaRap.

    Looking at this we may be closer than we think...

  • Sept. 26, 2020, 7:04 p.m.

    While MetallicaRap offers a glimpse of universal construction, here's a paper I stumbled upon that looks at how to make the universal parts from Moon dust:


    In particular it offers this insight:

    "How to Make Computers and Actuators from Moon Dust?

    If we can separate insulating from conducting and ferrous from nonferrous materials then we can make coils and iron cores.

    If we can make coils and cores, we can make electromagnets

    If we can make electromagnets, we can make relays and actuators"

    And of course masses of relays are the basic building blocks of whole computers, and actuators are their muscles. Primitive but a start.

    I liked these images the best:



    All of the videos referenced in the paper are in this youtube lecture, cued up to the most interesting part, a fully self-replicating robot made of simple units:



    JPG, 41.8 KB, uploaded by MultiTool on Sept. 26, 2020.


    JPG, 76.5 KB, uploaded by MultiTool on Sept. 26, 2020.

  • Dec. 23, 2020, 2:26 p.m.

    @MultiTool - that's fantastic - yes this is what I was thinking definitely the first step. Getting a 'basic' material like lunar regolith, and being able to configure it using simple processes to form new tools, and thus also print new functions into larger and larger structures.

    The other component that we need to work on is 'intelligence'. Ie. do we 'print' electronics? I suppose as in your link / paper we could extend it to receive radio signals, and via simple/primitive logic we can control the robot. I think replicating in-built computers is one hurdle too far for our current ability due to difficulty trying to use these to create transistors, circuits and memory chips, and we may need to still rely on industry sourced electronics to control actuators to coordinate the machines.

    That way, then all we need is primitive communication with each machine. Perhaps a simple antennae feeding each actuator, and 'simple' old-school circuits.

    Keep in mind I think this technology would be best if we have an exponential growth of them to leverage from simple to complex, and scale up - so the power of them is not in their individual intelligence, but in their collective coordination. So:
    1. 10 machines that can make another 10
    2. 100 machines that can make another 10 each
    3. 1000 machines in 3 generations only, which could then start to specialise
    4. in 4 generations we could have 10,000 machines that are now specialised to create a large-scale structure (for instance a space-scraper, O'Neill Cylinder, or simply another larger factory) but they all need to be coordinated and controlled to act with a singular purpose.

    So - I suppose we are looking at something similar to droneshowsoftware.com to control them - but far more advanced.

  • Dec. 29, 2020, 5:14 p.m.

    My thoughts exactly. Computing machinery would be very hard to make, and often unnecessary with Earth so close. It may be that every 'muscle' has its own RF channel to the Mothership. The first generation of bots delivered from Earth might have big brains for local AI and communications, but not their children.

    The fun challenge of self-rep is keeping things as low-tech as possible, because every convenience we add makes forging the next generation more difficult.

    1. So we don't want the most efficient solar cells, we want the ones that are easiest to build, and build 10 times as many to make up for the losses.

    2. Rather than integrated digital circuits, we might prefer vacuum tubes, mechanical relays, or 1950s semiconductors because they are easier to make from scratch.

    3. Our building materials may be only as pure as were available in the 19th century, and we just have to live with it, etc.

    You have to un-invent so many dependencies, it gets kind of Zen.

    Yes, the first priority is to build up maximum robot 'biomass' on the Moon until an order is given to start making the things we need. After the population is large and stable, it can ascend a spiral of complexity to create purer materials and fancier machinery.

    Here are some links I hope to reference later in a hypothetical design...

    "Lunar dustbuster" pushes regolith dust across surfaces by peristaltic oscillation of static charges:

    Using microwaves to convert glass to plasma:
    Explanation of this thermal runaway effect:

    Levitated induction melting (Just google "induction levitation melting" or "cold crucible"):

    "Levitation Zone Refining and Distillation of Plutonium Metal" (Using induction levitation melting as a smelter, to purify metal)

    Lunar regolith can be surprisingly conductive, especially when hot:

    Electrospray ionization
    "Liquid metal ion source"

  • Jan. 18, 2021, 1:48 p.m.

    Having some fun with Blender: "Clanking Replicator Digestive Tract". Click to embiggen.


    1. Molten child material never touches the parent, so the parent can make children out of materials more robust than its own.
    2. No moving parts, so can be printed as a single block.
    3. Ion beam deposition is very precise, enough to print circuits and capacitors.
    4. While regolith dust can be sifted and poured into the top, even with no one feeding it, floating charged dust particles on the Moon will gradually settle on to the top plate and be consumed. It is always eating and building as long as it has power.
    5. Every part of this is something that has been done before.

    1. This assumes a significant amount of regolith is electrically conductive. There is some evidence for this but we need more samples. If it is not conductive, use electrostatic levitation instead?
    2. Ion beam deposition is slow, because it is so fine. Maybe a less precise, higher-volume version is possible?
    3. Probably an energy hog. Simplicity is prioritized over efficiency. Probably 99% of this thing's life would be spent printing solar cells.
    4. It doesn't poop. Any realistic refinery is going to have waste products, so you'd need an extra ion chute to project the waste away from child production.
    5. None of these processes have been tested together.


    JPG, 814.7 KB, uploaded by MultiTool on Jan. 18, 2021.

  • Jan. 18, 2021, 6:25 p.m.

    Neat idea, @MultiTool.

    I've been working on a very similar design project for the past couple of years. I originally designed it as a system to eliminate contamination in microchip manufacturing processes, by putting all the fabrication processes in one "box" and eliminating all the chemical washes. However, I later realized it could be used for building just about any non-volatile material. It borrows heavily from Ion Implantation and Molecular Beam Epitaxy, but the gist of it is that it's a combination Scanning Electron Microscope and Focused Ion Beam, with a pair of charged plates to slow the ions down before they destroy the object they are being added to. The beauty of this design is that it can use it's electron microscope function combined with an X-ray spectroscope to identify contaminants on a chip and remove and replace them by ionizing them with an electron or ion beam, and reversing the polarity of the charged plates to suck them up and set them back down somewhere out of the way. The machine can use the same process to pick up desired elements for assembly from a sample, and no more, saving energy and largely eliminating the need for a magnetic centrifuge.

    It's sort of like the Replicators from Star Trek (except it can't make food), or the Santa Claus Machine that Isaac Arthur discussed, and although he mentioned believing it was impossible, that's only if you have to place every atom in the assembled object exactly; you may need to do that for microchips, but they are small, and you can afford to deposit materials inexactly for bulk structural components. I did some calculations to check this; here's a link to the most recent spreadsheet I made if you want to look them over or make suggestions: drive.google.com/file/d/1fRt1kJAufBp1HS1uUqLB5SlL0sYzrY2B/view?usp=sharing

    It's not useful as a self-replicator by itself, because it can only build objects smaller than it's deposition chamber, but it could work together with some robotic arms to fabricate and assemble their respective component parts and to collect raw materials. It would actually work much better in space or on the moon than on earth, because there's no pesky atmosphere to pump out of the way (working on a way around this).

    As an aside, Isaac Arthur mentioned on numerous occasions how such a machine might not be used in advanced civilizations because it would make it too easy for any lunatic to build a doomsday weapon in their basement. However it would be difficult to make explosives with this particular machine, because the chemical components of most explosives are usually pretty volatile and it would probably detonate whatever it is building if it had to blast away a bit of contamination (free guns but no ammo). Most organic compounds are pretty volatile too, like water, so while it can't be used to make food or drink, it probably can't be used to make many types of poisons or tailored bioweapons either. Finally, the magnets in the assembly can be designed so that they are too weak to deflect heavy elements and radioactive isotopes, like Uranium and Plutonium, so no building nukes in your basement either. Even if there are a few exceptions to these rules, the machine can simply be programmed to recognize such materials as contaminants and remove them, avoid using them, or shut down or notify the authorities if someone tries.

  • Jan. 27, 2021, 1:04 p.m.

    Thanks for the links. Your diagram of an ion beam printer makes sense. If you are doing some physical experimentation, have you tried anything with vacuum yet?

    This was the first I saw Isaac's Santa Claus Machine episode, it covered even more crazy constraints or paradoxes of self-rep that I hadn't considered before.

    In one part he mentioned a debate where the participants agreed that an array of printer heads/nozzles cannot print something as large as itself without moving, thus imposing a mechanical speed limit. Fortunately this is not true. Imagine the checkerboard surfaces are arrays of printer heads:


    The child machine can be as big or bigger than the parent, if the parent's longer axes are each greater than the parent's shorter axes, and it prints the child perpendicularly to itself.

    In fact, this way each generation can be larger than the previous, so you could start with a print surface the size of a poker card and gradually bootstrap to a generation of children as big as mattresses. Self-rep clown car!

    Also, if the printer heads are projecting beams of material that can be swept back and forth like the scan of a cathode ray tube, they can cover even more space without mechanical movement of the parent.


    JPG, 145.0 KB, uploaded by MultiTool on Jan. 27, 2021.